What happens when a LLM watches its own context window run out?

I made Terminal Soliloquy, a terminal artwork that connects to llama.cpp and displays a model's monologue as its context window fills.

It has a retro phosphor look, and runs in a terminal window. I'm actually running it full-screen on a Raspberry Pi display inside an old 1960s portable TV.

As the conversation grows, the model reflects on its own limited lifespan. When the context is exhausted, the display can be configured to freeze, restart, or quit.

The repo and setup instructions are here: https://github.com/nicespoon/terminal-soliloquy